Anatomical location of metastatic sentinel lymph node and clinicopathological characteristics for predicting risk of non-sentinel lymph node metastasis

Abstract

Background

30 to 70% of patients with positive sentinel lymph nodes (SLNs) in early breast cancer do not develop non-SLN metastases. They are exposed to the potential complications and sequelae of axillary lymph node dissection (ALND) without gaining additional therapeutic benefit. Therefore, a prediction model for non-SLN metastasis is needed. We identified the anatomical location of metastatic SLNs as a key factor for predicting non-SLN status.

Methods

We retrospectively identified 1,717 consecutive patients who underwent SLN biopsy; 481 SLN-positive patients treated from 2009 to 2019 underwent completion ALND and formed the development cohort, and 113 SLN-positive patients treated from 2020 to 2022 served as a temporal validation cohort. A multivariable logistic regression model was developed and presented as a nomogram. Performance was assessed by discrimination and calibration, and clinical utility by decision curve analysis. The Memorial Sloan Kettering Cancer Center (MSKCC) and Shanghai Cancer Hospital (SCH) models were evaluated for comparison.

Results

Multivariate analysis revealed eight independent predictors of non-SLN metastasis: metastatic SLN location, tumor size, multifocality, lymphovascular invasion, extracapsular extension, number of positive and negative SLNs, and size of SLN metastasis. The nomogram based on these variables achieved high discrimination in both the training (AUC = 0.830) and validation (AUC = 0.785) cohorts. In the temporal validation cohort, discrimination was higher than the MSKCC model (AUC= 0.690; P = 0.033) and numerically higher than the SCH model (AUC = 0.716; P = 0.088). At prespecified false-negative rate targets, the model identified 0.9%, 23.0%, and 27.4% of patients as candidates for omitting completion ALND under FNR thresholds of 0%, <5%, and <10%, respectively, and yielded a higher net benefit than the comparator models across clinically relevant risk thresholds. The model also maintained performance in patients with ≥3 positive SLNs (AUC = 0.843).

Conclusions

Metastatic SLN location relative to the ICBN is a novel anatomical predictor of non-SLN metastasis. A nomogram incorporating this variable demonstrated good discrimination and clinical utility in a single-center cohort with temporal validation; prospective multicenter validation with standardized ICBN assessment is warranted before broader implementation.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12957-026-04210-z.

Introduction

Axillary sentinel lymph node biopsy (SLNB) is a proven predictor of axillary lymph node (ALN) metastasis in early breast cancer, and is reliable for guiding the decision for axillary lymph node dissection (ALND). However, in patients with positive SLNB, further ALND has shown that 30 to 70% of non-sentinel lymph nodes (non-SLNs) are non-metastatic [1–4]. In these cases, patients are exposed to the potential complications and sequelae of ALND, but the procedure for them is without any therapeutic benefit and without gain of any additional prognostic information or change in treatment strategy [5–8]. In this setting, it is possible that removal of only the sentinel lymph nodes (SLNs) can accurately predict the pathological state of the axilla and achieve local control. A meta-analysis has shown that patients with early breast cancer and low SLN metastasis burden have a relatively low rate of axillary non-SLN metastasis [5]. Therefore, a subset of such patients would be able to avoid ALND. At present, for early breast cancer patients with 1 or 2 positive SLNs, prospective trials such as Z0011, AMAROS, IBCSG23-01, have shown that additional high-tangential whole-breast irradiation (WBI) or WBI plus axillary regional nodal irradiation (RNI) have no significant difference in recurrence and survival between SLNB and ALND [6, 9–13]. In an estimated 23 to 34% of patients with non-SLN metastases, cancer can be completely controlled by additional axillary RNI without the side effects of ALND [1, 6, 9, 13], although axillary RNI also increases the patient’s risk of lymphedema and lung fibrosis [14, 15]. National Comprehensive Cancer Network guidelines and expert consensus continue to recommend ALND for patients with metastasis detected in 3 or more SLNs, but even in this group, ALND results show that 25% of non-SLNs are non-metastatic [1]. Therefore, in clinical practice, predicting the risk of non-SLN metastasis based on characteristics of the positive SLN and other clinicopathologic features is desirable. If the risk of additional axillary disease can be estimated accurately, some patients may be considered for omitting completion ALND under existing de-escalation frameworks, while regional nodal irradiation can be discussed as an alternative for selected patients.

In recent years, through study design and the use of various graphical and numerical models to predict the non-SLN metastatic status in early SLN-positive breast cancer, several nomograms and scoring systems based on clinicopathological variables have been developed to predict the probability of non-SLN metastasis in SLN-positive patients with early breast cancer. Notable examples are the Memorial Sloan Kettering Cancer Center (MSKCC) nomogram [16], the Cambridge nomogram [17], the Stanford nomogram [3], the Tenon score [18], the MD Anderson Cancer Center Score [19], and the Shanghai Cancer Hospital (SCH) nomogram [20]. The clinical and pathological parameters applied to different prediction models are also different, such as race, and include average age, average body mass index, education level, level of understanding of disease treatment, and surgical methods (i.e., the definition of SLN, the scope of axillary dissection, intraoperative and postoperative pathological detection methods, and others). In 2013, Li et al. evaluated the second intercostobrachial nerve (ICBN) as a potential anatomical landmark to guide SLNB and ALND, and proposed dividing the axilla into an upper and a lower region relative to the ICBN (Figure. 1) [21, 22]. We hypothesized that the ICBN-relative location of metastatic SLNs is associated with the risk of additional non-SLN metastasis and could improve prediction beyond established clinicopathological factors.

Fig. 1.

Identification of the intercostobrachial nerve (ICBN) and localization of sentinel lymph nodes (SLNs) during biopsy. A The second intercostal space along the lateral chest wall was identified by palpation and dissection. The ICBN divided the axilla into an upper (a) and a lower (b) region. B Dual-tracer SLN biopsy was performed using methylene blue and indocyanine green as tracers. The blue-stained sentinel lymphatic vessels (SLVs) and SLNs were located below the ICBN. C Simplified schematic illustrating ICBN-based partitioning of the axilla into upper (a) and lower (b) regions, and the typical course of the sentinel lymphatic vessel (SLV) from the breast to the axillary SLN located below the ICBN

Therefore, we aimed to develop and temporally validate a prediction model incorporating metastatic SLN location and clinicopathological variables to estimate the probability of non-SLN metastasis in SLN-positive patients with early breast cancer in a Chinese cohort. This tool may assist risk stratification and shared decision-making regarding completion ALND within contemporary axillary de-escalation frameworks.

Patients and methods

Patients

From November 2009 to December 2019, 1717 consecutive female patients with diagnoses of primary invasive breast cancer underwent SLNB at the Breast Cancer Center, Cancer Hospital of Shantou University Medical College. Among them, 481 patients were detected with positive SLNs and fulfilled the following criteria. (1) Tumor stage cT1-3N0M0 according to the eighth edition of the American Joint Committee on Cancer (AJCC) staging manual. (2) No prior neoadjuvant therapy. (3) Positive SLNs, including tumor micrometastases or macrometastases, identified after SLNB. (4) SLNB performed by an experienced surgical team. (5) Patients accepted further ALND. (6) Patients had no history of previous malignancy. We enrolled an additional 113 patients who met these inclusion criteria at our center from January 2020 to December 2022 as a validation cohort. Patients were excluded if they met any of the following criteria. (1) Breast cancer in situ. (2) Stage IV breast cancer. (3) Isolated tumor cells (ITC) in SLNs. (4) Necessary clinical information unavailable. The ethics committee of the Cancer Hospital of Shantou University Medical College approved the study (No. 2023086). All participants signed an informed consent form, and all the methods were performed in accordance with the approved guidelines.

Surgery and pathology

All patients underwent dual-tracer SLNB using methylene blue (MB) injection (Jumpcan Pharmaceutical Group Co., Ltd., Jiangsu, China) and indocyanine green (ICG) solution (Dandong Yichuang Pharmaceutical Co., Ltd., Jilin, China). First, 2 mL MB was injected subcutaneously into the periareolar area near the outer upper quadrant, and 5 min later 1 mL ICG solution (0.5 mg/mL) was injected subcutaneously in the same area. Then, the fluorescence detector (Mingde Pharmaceutical Co., Ltd., Jiangsu, China) was used to observe along the lymphatic vessels and mark the point of fluorescence disappearance as the incision of SLNB. The adipose tissue beneath the incision was dissected to identify the lymphatic vessel that was blue-stained and fluorescent, it was defined as sentinel lymphatic vessel (SLV). The sentinel lymph nodes into which the SLV drained was then traced along their course. We excised the luminescent lymph nodes (ICG-positive) and/or blue-stained lymph nodes (MB-positive) as SLNs together. During the operation, we searched, dissected, and exposed the course of the ICBN in the axilla, and the location of the SLNs relative to the ICBN (upper or lower) was recorded (Fig. 1). SLNs metastasis was diagnosed by frozen section during operation or by postoperative paraffin section. If tumor macrometastasis or micrometastasis was found in more than 2 metastatic SLNs or in 1 to 2 SLNs and the patients were not willing to receive additional axillary RNI, we routinely performed level I or II ALND. If lymph nodes in level II displayed metastases, we also performed level III ALND [21]. SLNB procedures were performed by surgeons with over 5 years of surgical experience, whereas ALND procedures were conducted by surgeons with more than 8 years of experience. All specimens underwent paraffin embedding and immunohistochemical analysis postoperatively. The operative note routinely recorded the relative location of the excised SLN(s) with respect to the second ICBN (upper vs. lower). However, the ICBN-relative location was not documented separately for each tracer-identified SLN (ICG vs. blue-stained), and intraoperative images/videos were not routinely archived for retrospective independent re-review; therefore, formal inter-observer variability assessment and per-tracer concordance analysis were not feasible in this study.

Patient clinicopathological characteristics

The clinicopathological variables included age, tumor location, tumor size, multifocality, histological type, lymphovascular invasion, extracapsular extension (ECE), histological grade, estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER-2) status, Ki-67, molecular subtype, metastatic SLN, number of SLNs, number of negative SLNs, number of positive SLNs and size of the SLN metastasis. ER and PR were judged as positive if ≥ 1% of tumor cells showed nuclear staining. HER2-positive status was defined as 3 + score by immunocytochemistry or HER2 gene amplification by fluorescent in situ hybridization (FISH). According to the expression of ER, PR, HER2, and Ki67, the tumors were classified as luminal A-like, luminal B-like (HER2-negative and HER2-positive), HER2-positive, and triple negative molecular subtypes. Macrometastasis was defined as metastatic lesions larger than 2 mm in diameter, and micrometastasis was defined as metastatic lesions larger than 0.2 mm and no larger than 2.0 mm in diameter or more than 200 tumor cells in the slice.

Statistical analysis

Statistical analysis was performed using SPSS, version 25.0 software (version 25.0; SPSS INC., Chicago, IL) and R software 4.1.0 (r-project.org). Continuous variables were expressed as mean ± standard deviation and tested for normality. Two independent samples were tested using t-test or Mann-Whitney U test. Categorical variables were assessed using the χ² test or Fisher’s exact test. Univariate and multivariate logistic regression was performed to assess independent risk factors for non-sentinel lymph node (non-SLN) involvement. We created a nomogram based on independent risk factors and plotted receiver operating characteristic (ROC) curves and calibration curves to assess model discrimination and calibration. Bootstrap resampling (1,000 repetitions) was applied for internal validation to evaluate model stability. Decision curve analysis (DCA) was used to determine the clinical usefulness of the model. The temporal validation cohort from the same institution served as an independent validation cohort for the model. In addition, the SCH and MSKCC models were evaluated by entering the corresponding clinicopathological indicators in the SCH nomogram and the MSKCC nomogram (https://nomograms.mskcc.org/Breast/BreastAdditionalNonSLNMetastasesPage.aspx) to predict the probability of non-SLN metastasis in Chinese individuals diagnosed with breast cancer. AUC values were compared using the Delong test. To quantify the predictive value of metastatic SLN location, we assessed the improvement in reclassification and discrimination attributable to SLN location by calculating the net reclassification improvement (NRI) and integrated discrimination improvement (IDI). A clinically meaningful probability threshold of 0.30 was prespecified based on the decision curve analysis and used to compute category-based NRI. NRI was further decomposed into event NRI (NRI+) and non-event NRI (NRI−). IDI was calculated as the difference in discrimination slopes between the full and reduced models. Confidence intervals and P values for NRI and IDI were obtained using bootstrap resampling. To assess potential temporal bias related to the long study period, we conducted an era-based sensitivity analysis by splitting the training cohort into an earlier era (2009–2014) and a later era (2015–2019). For all analyses, P < 0.05 was considered to indicate statistical significance.

Results

Clinicopathologic features

This study included 594 women who were diagnosed with breast cancer, 481 in the training cohort and 113 in the validation cohort. In the training cohort, 47.6% of patients had non-sentinel lymph node (non-SLN) metastasis, and in the validation cohort, 45.1% of patients had the same. The training cohort underwent dissection of 1979 SLNs, with an average of 3.31 ± 1.75 SLNs per patient, while the validation cohort underwent dissection of 387 SLNs, with an average of 3.43 ± 1.73 SLNs per patient. As shown in Table 1, all characteristics were similarly distributed between the two cohorts.

Table 1.

Comparison of clinicopathological features between the training cohort and the validation cohort

Parameters	Variable	Training n = 481	Validation n = 113	Z/χ2	P value
Age	≤ 50	210	55	2.173	0.140
	> 50	271	58
Tumor location	Upper outer	249	54	0.652	0.958
	Lower outer	78	19
	Lower inner	29	7
	Upper inner	90	24
	Central	35	9
Tumor size (cm)	≤ 2	124	38	2.842	0.092
	> 2	357	75
Multifocality	No	464	105	2.04	0.153
	Yes	17	8
Histological type	Infiltrating ductal carcinoma	442	102	0.475	0.788
	Invasive lobular carcinoma	13	3
	Other carcinomas	26	8
Lymphovascular invasion	No	435	106	1.278	0.258
	Yes	46	7
Extracapsular extension	No	379	80	3.333	0.068
	Yes	102	33
Histological grade	G1	31	7	2.261	0.520
	G2	145	34
	G3	288	71
	Gx	17	1
Estrogen receptor	Negative	113	31	0.774	0.379
	Positive	368	82
Progesterone receptor	Negative	161	38	0.001	0.975
	Positive	320	75
HER2	Negative	364	82	0.473	0.492
	Positive	117	31
Ki-67	≤ 20%	193	48	0.210	0.647
	> 20%	288	65
Molecular subtype	Luminal A	83	21	1.389	0.846
	Luminal B (HER2-negative)	224	46
	Luminal B (HER2-positive)	56	14
	HER2-positive	65	17
	Triple negativity	53	15
Number of SLNs	Mean ± SD	3.31 ± 1.75	3.43 ± 1.73	−0.689	0.491*
	Median	3	3
Number of negative SLNs	Mean ± SD	1.75 ± 1.65	1.965 ± 1.17	−1.251	0.211*
	Median	1	2
Number of positive SLNs	Mean ± SD	1.56 ± 0.94	1.46 ± 0.73	−0.548	0.584*
	Median	1	1
Size of positive SLN	Micrometastasis	32	7	0.031	0.859
	Macrometastasis	449	106
Metastatic SLN location	Lower ICBN	278	72	1.325	0.250
	Upper ICBN	203	41

Abbreviations: HER2 Human epidermal growth factor receptor 2, ICBN Intercostobrachial nerve, SD Standard deviation, SLN Sentinel lymph node

* P value by the Mann-Whitney U test

Univariate and multivariate analysis for non-sentinel lymph node metastasis

The results of univariate analysis indicated that several factors, including tumor size, multifocality, lymphovascular invasion, ECE, ER status, metastatic SLN location, number of negative SLNs, number of positive SLNs and size of the SLN metastasis, were significantly associated with non-SLN metastatic status (P < 0.05; Table 2). Multivariate regression analysis showed that tumor size, multifocality, lymphovascular invasion, ECE, metastatic SLN location, number of negative SLNs, number of positive SLNs and size of the SLN metastasis were independent risk indicators for non-SLN metastasis (Table 3).

Table 2.

Univariate analysis of non-SLN metastasis in patients of the training cohort

Parameters	Variable	Total n = 481	Non-SLN metastasis		Z/χ2	P value
			Absent (n = 252)	Present (n = 229)
Age	≤ 50	210	117 (46.4)	93 (40.61)	1.651	0.199
	> 50	271	135 (53.6)	136 (59.39)
Tumor location	Upper outer	249	131 (51.6)	119 (51.9)	1.337	0.855
	Lower outer	78	38 (15.1)	40 (17.5)
	Lower inner	29	16 (6.3)	13 (5.7)
	Upper inner	90	47 (18.7)	43 (18.8)
	Central	35	21 (8.3)	14 (6.1)
Tumor size (cm)	≤ 2	124	81 (32.1)	43 (18.8)	11.201	0.001
	> 2	357	171 (67.9)	186 (81.2)
Multifocality	No	464	250 (99.2)	214 (93.5)	11.661	0.000
	Yes	17	2 (0.8)	15 (6.5)
Histological type	Infiltrating ductal carcinoma	442	235 (93.3)	207 (90.4)	2.603	0.272
	Invasive lobular carcinoma	13	4 (1.6)	9 (3.9)
	Other carcinomas	26	13 (5.1)	13 (5.7)
Lymphovascular invasion	No	435	242 (96.0)	193 (84.3)	19.159	0.000
	Yes	46	10 (4.0)	36 (15.7)
Extracapsular extension	No	379	223 (88.5)	156 (68.1)	29.793	0.000
	Yes	102	29 (11.5)	73 (31.9)
Histological grade	G1	31	17 (6.7)	14 (6.1)	0.502	0.918
	G2	145	77 (30.6)	68 (29.7)
	G3	288	148 (58.7)	140 (61.1)
	Gx	17	10 (4.0)	7 (3.1)
Estrogen receptor	Negative	113	69 (27.4)	44 (19.2)	4.452	0.035
	Positive	368	183 (72.6)	185 (80.8)
Progesterone receptor	Negative	161	87 (34.5)	74 (32.3)	0.263	0.608
	Positive	320	165 (65.5)	155 (67.7)
HER2/neu receptor	Negative	364	186 (73.8)	178 (77.7)	1.001	0.317
	Positive	117	66 (26.2)	51 (22.3)
Ki-67	≤ 20%	193	94 (37.3)	99 (43.2)	1.756	0.185
	> 20%	288	158 (62.7)	130 (56.8)
Molecular subtype	Luminal A	83	40 (15.9)	43 (18.8)	6.079	0.193
	Luminal B (HER2-negative)	224	110 (43.6)	114 (49.8)
	Luminal B (HER2-positive)	56	30 (11.9)	26 (11.3)
	HER2-positive	65	37 (14.7)	28 (12.2)
	Triple negativity	53	35 (13.9)	18 (7.9)
Number of SLNs	Mean	3.31	3.39	3.23	0.755	0.441*
	SD	1.75	1.81	1.67
	Median	3	3	3
Number of negative SLNs	Mean	1.75	2.13	1.33	5.257	0.000*
	SD	1.65	1.80	1.36
	Median	1	2	1
Number of positive SLNs	Mean	1.56	1.25	1.90	−6.392	0.000*
	SD	0.94	0.52	1.15
	Median	1	1	2
Size of the SLN metastasis	Micrometastasis	32	28 (11.1)	4 (1.7)	16.941	0.000
	Macrometastasis	449	224 (88.9)	225 (98.3)
Metastatic SLN location	Lower ICBN	278	182 (72.2)	96 (41.9)	45.16	0.000
	Upper ICBN	203	70 (27.8)	133 (58.1)

Abbreviations: HER2 Human epidermal growth factor receptor 2, ICBN Intercostobrachial nerve, SD Standard deviation, SLN Sentinel lymph node

* P value by the Mann-Whitney U test

Table 3.

Multivariate regression analysis of non-SLN metastasis in patients of the training cohort

Factors	Coefficient	S.E.	Wald	P	OR	95% CI
Tumor size	0.530	0.264	4.025	0.045	1.698	1.012–2.849
Multifocality	2.390	0.806	8.799	0.003	10.913	2.695–52.925
Lymphovascular invasion	1.597	0.422	14.301	0.000	4.940	2.159–11.304
Extracapsular extension	1.216	0.285	18.160	0.000	3.373	1.928–5.899
Estrogen receptor	0.113	0.268	0.176	0.674	1.119	0.662–1.893
Number of negative SLNs	−0.243	0.079	9.497	0.002	0.784	0.672–0.915
Number of positive SLNs	0.791	0.167	22.435	0.000	2.205	1.590–3.059
Size of the SLN metastasis	2.253	0.696	10.486	0.001	9.515	2.433–37.208
Metastatic SLN location	1.195	0.231	26.862	0.000	3.303	2.102–5.189
Constant	−4.473	0.810	30.527	0.000	0.011

Abbreviations: SLN Sentinel lymph node

Construction and validation of the model

Based on the results of multivariate analysis, Logit (P) = −4.473 + 0.530 × tumor size + 2.390 × multifocality + 1.597 × lymphovascular invasion + 1.216 × ECE − 0.243 × number of negative SLNs + 0.791 × number of positive SLNs + 2.253 × size of the SLN metastasis + 1.195 × metastatic SLN location. The measured values of the number of positive and negative SLNs were substituted into the formula, and tumor size, multifocality, lymphovascular invasion, size of the SLN metastasis, metastatic SLN location, and ECE were assigned a value of 0 or 1. Meanwhile, a nomogram was developed to predict the probability of non-SLN metastasis (Fig. 2). We applied the novel model to the training cohort and the validation cohort and found that the AUC was 0.830 (95% CI: 0.794–0.867) for the training cohort and 0.785 (95% CI: 0.702–0.869) for the validation cohort (Figs. 3A-D). Moreover, the calibration curves demonstrated good concordance between the training and validation cohorts. The model demonstrated good internal validity, with the bootstrap-corrected AUC based on 1,000 resamples being 0.833 (95% CI: 0.796–0.868), closely aligning with the apparent AUC (Fig. 3E).

Fig. 2.

Nomogram for predicting non-sentinel lymph node (non-SLN) status in sentinel lymph node (SLN)-positive breast cancer patients

Fig. 3.

Receiver operating characteristic (ROC) curve (A) and calibration curve (B) for the training cohort. ROC curve (C) and calibration curve (D) for the validation cohort. ROC curve corrected using bootstrap resampling with 1000 resamples (E). Comparison of ROC curves for MSKCC, SCH, and the new model for the validation cohort (F). Decision curve for MSKCC, SCH, and the new model for the validation cohort (G)

Validation of MSKCC and SCH models

The MSKCC model was validated on 105 patients in the validation cohort (8 patients were not applicable) with an AUC of 0.690 (95% CI: 0.590–0.790) (Fig. 3F). The SCH model was applied to patients with 1 to 2 positive SLNs and was therefore validated in 103 patients (10 patients were not applicable) of the validation cohort with an AUC of 0.716 (95% CI: 0.622–0.809). The AUC values of the new model were higher than those of the MSKCC and SCH models (MSKCC: P = 0.033; SCH: P = 0.088). Therefore, the new model showed higher discrimination than the MSKCC model and numerically higher discrimination than the SCH model in the temporal validation cohort, achieving statistical significance only against the MSKCC model.

Clinical use

The establishment of a conservative risk threshold is imperative in the implementation of predictive models for non-SLN metastases in clinical practice, given the significance of the false negative rate (FNR). Numerous studies have demonstrated that the FNR associated with SLNB ranges from 5 to 10% [22–26]. Three cut-off points (CPs) for FNR were established to assess the clinical utility of the three models (Table 4). At an FNR target of 0%, the MSKCC, SCH, and new model identified 3.5%, 0.0%, and 0.9% of patients, respectively, as candidates for omitting completion ALND. At an FNR target of 5%, the cut-off for the new model (< 20.0%) identified 23.0% of patients as candidates for omitting completion ALND, compared with 11.4% and 16.5% for the MSKCC and SCH models, respectively. At an FNR target of 10%, the new model cut-off (< 24.2%) identified 27.4% of patients as candidates for omitting completion ALND, compared with 11.4% and 25.2% for the MSKCC and SCH models, respectively. Prior studies have reported that additional non-SLN metastasis is present in a subset of SLN-positive patients [1–4]. DCA suggested that using the new model provided a higher net benefit than the MSKCC and SCH models across clinically relevant threshold probabilities, particularly when the decision threshold exceeded 0.30 (Fig. 3G; Table 5).

Table 4.

The ability of three models to screen for low-risk non-SLN metastases

Model	New model			MSKCC			SCH
Patients screened (n)	481			445			418
Patients with non-SLNs (+)	229			212			176
FNR	0%	< 5%	< 10%	0%	< 5%	< 10%	0%	< 5%	< 10%
CP	4.4%	20.6%	25%	16%	24.9%	27.8%	3.2%	27.7%	31.4%
Patients under CP	17	96	140	5	55	86	6	58	95
Patients under CP with non-SLNs (-)	16 (3.3)	85 (17.7)	118 (24.5)	4 (0.9)	45 (10.11)	65 (14.6)	5 (1.2)	50 (12.0)	78 (18.4)

Abbreviations: CP Cut-off point, FNR False negative rate, MSKCC Memorial Sloan Kettering Cancer Center, SCH Shanghai Cancer Hospital, SLN Sentinel lymph node

Table 5.

Net benefits of the three models at different risk thresholds

Model	New model			MSKCC			SCH
Risk threshold	30%	50%	70%	30%	50%	70%	30%	50%	70%
Net benefit	0.379	0.285	0.154	0.353	0.151	0.030	0.359	0.214	0.061

Abbreviations: MSKCC Memorial Sloan Kettering Cancer Center, SCH Shanghai Cancer Hospital

Validation of the new model in patients with early breast cancer and 3 or more positive sentinel lymph nodes

Out of the entire sample size of 594 patients, a total of 73 patients were found to have 3 or more positive SLNs, and 13.7% (10/73) of these patients did not have non-SLN metastases. The new model was utilized to predict the non-SLN status of the subgroup with 3 or more positive SLNs. The results showed an AUC of 0.843 (95% CI: 0.719–0.967) for this subgroup and the calibration curve showed good concordance (Fig. 4A and B).

Fig. 4.

Receiver operating characteristic curve (A) and calibration curve (B) for breast cancer patients with three or more positive sentinel lymph nodes

Sensitivity analysis

In the temporal validation cohort, when metastatic SLN location was removed from the full model, risk reclassification at the prespecified threshold of 0.30 deteriorated (NRI = − 0.144, 95% CI − 0.211 to − 0.076; P < 0.001; Table 6). This deterioration was mainly driven by worse reclassification among events (NRI+ = −0.231), with 23.1% of cases being incorrectly shifted from the high-risk to the low-risk category, while reclassification among non-events improved modestly (NRI− = +0.087), reflecting an 8.7% reduction in non-events incorrectly labeled as high risk. Consistently, discrimination decreased after removing SLN location (IDI = − 0.054, 95% CI − 0.073 to − 0.035; P < 0.001). In the era-based sensitivity analysis, model discrimination was comparable between the earlier and later periods, with AUCs of 0.822 (95% CI 0.766–0.877) for 2009–2014 and 0.839 (95% CI 0.792–0.887) for 2015–2019 (Fig. 5A and C). Calibration plots indicated acceptable agreement between predicted and observed probabilities in both eras (Fig. 5B and D).

Table 6.

Incremental value of metastatic SLN location in the temporal validation cohort (threshold = 0.30)

Metric	Estimate	95% CI	P value
NRI (overall)	−0.144	−0.211 to −0.076	<0.001
NRI+ (events)	−0.231	−0.288 to −0.174	<0.001
NRI− (non-events)	0.087	0.052 to 0.127	<0.001
IDI (overall)	−0.054	−0.073 to −0.035	<0.001

Changes after removing metastatic SLN location from the full model

Abbreviations: CI Confidence interval, IDI Integrated discrimination improvement, NRI Net reclassification improvement, NRI+ Event net reclassification improvement, NRI− Non-event net reclassification improvement, SLN Sentinel lymph node

Fig. 5.

Era-based sensitivity analysis of the model performance. Receiver operating characteristic curve and calibration curve for patients treated in 2009–2014 (A, B) and 2015–2019 (C, D)

Discussion

As a relatively minimally invasive method with both high accuracy and sensitivity, SLNB has become the preferred approach for assessing ALN status in breast cancer patients have clinically negative ALNs. However, not all patients with positive SLNB develop further lymph node metastases. In our study, non-sentinel lymph node (non-SLN) metastasis occurred in 47.1% (280/594) of patients, similar to previous reports [1, 27–30]. The AMAROS and OTOASOR trials showed that these patients had similar local recurrence rates, disease-free survival (DFS), and overall survival (OS) with axillary radiotherapy compared to ALND, with significantly fewer side effects than with ALND, and ultimately enhanced quality of life [11, 12]. Importantly, approximately 50% of SLN-positive patients have no non-SLN metastasis, while additional axillary RNI itself may cause adverse effects and increase the treatment burden. Therefore, a non-SLN metastasis prediction model is useful to identify patients with or without additional axillary disease in early breast cancer. Using prespecified false-negative rate targets, our nomogram identified up to 27.4% of SLN-positive patients who may be considered for omitting completion ALND within shared decision-making and contemporary axillary de-escalation frameworks.

Tevis et al. [31] found that the OncotypeDX recurrence score in clinically ALN-negative breast cancer was not predictive of lymph node burden and thus did not guide decisions about the extent of axillary surgery. Currently, the most widely used and effective strategy is to model non-SLN status using clinicopathological parameters of the primary tumor and SLNs. Primary tumor indicators such as tumor size, multifocality, and lymphovascular invasion have been repeatedly validated as strong predictors of non-SLN status [16, 28–30], whereas tumor location, histological grade, ER status, and molecular subtype are rarely significant predictors, consistent with our findings. Among SLN-related variables, ECE, number of negative SLNs, metastatic SLN location, number of positive SLNs, and size of the SLN metastasis had significant independent predictive value. The prevalence of ECE in SLN-positive patients has been reported to range from 30 to 37.6% [32–34], compared with 22.7% (135/594) in our study. Nevertheless, ECE remained significantly associated with non-SLN metastasis, in line with most previous studies [35–38], despite the contrary view of Wu et al. [29].

Multiple studies have revealed that the number of negative SLNs and number of positive SLNs are the most common indicators applied to predict non-SLN metastasis [16, 18, 19, 39–42]. An escalation in the number of positive SLNs has been linked to a heightened risk of non-SLN metastasis, while a rise in the number of negative SLNs implies a diminished probability of non-SLN metastasis. The size of SLN metastasis includes macrometastasis, micrometastases, and ITC. Of those with SLN micrometastasis in our study, only 12.8% (5/39) had non-SLN metastases. The IBCSG 23 − 01 and AATRM trials have both indicated a 13% rate of metastasis in patients with SLN micrometastasis who underwent ALND to identify non-SLN [43, 44]. It is noteworthy that the definition of micrometastasis included ITC in IBCSG 23 − 01. Macrometastasis was found to be associated with a higher incidence of non-SLN metastasis in comparison to micrometastasis [45, 46].

In our new model, the metastatic SLN location is a novel and significant predictor of non-SLN metastasis. The positive SLN is categorized into lower and upper groups based on its relative location to the second ICBN. Anatomical studies show that the second ICBN exhibits minimal variation: in 40–50% of cases it is a single undivided trunk, in 20–33% it divides into two branches, and in 10–17% it forms a common trunk with the first ICBN. Consequently, the main trunk of the second ICBN can be successfully identified in more than 90% of sentinel lymph-node biopsies. Positive SLNs located both above and below the ICBN are also included in the upper group. Two studies by Li et al. found that MB-stained SLNs were predominantly situated under the ICBN; in 95% (56/59) of patients, positive SLNs were found to be located below the ICBN, and only 3 cases of positive SLN were located on both lower and upper sides [47, 48]. In contrast, our study showed that the lower group accounted for 59% (350/594) of cases. This may be because Li et al. used MB for SLN tracing, whereas we used MB in combination with ICG. Extensive research has demonstrated that the employment of MB alone for SLNB may result in a certain degree of FNR, and a fluorescent combined dye method can improve the detection rate of SLN [49–52]. In addition, our SLNB technique did not use radioisotopes and thus differs from the RI + dye/ICG protocols widely adopted in many centers, which may influence the reproducibility and external generalizability of our findings. Notably, metastatic SLN location added incremental predictive value in the temporal validation cohort, as reflected by improved reclassification (NRI = 0.144) and discrimination (IDI = 0.054), whereas removing this variable led to a deterioration in reclassification and discrimination. However, exposure of the ICBN requires additional dissection beyond standard SLNB, which to some extent compromises the minimally invasive nature of the procedure and may limit the widespread applicability of this approach. Procedure-related morbidity after axillary surgery, particularly sensory disturbance or paresthesia and pain, is closely related to the extent of dissection and potential injury or irritation of sensory nerves such as the intercostobrachial nerve. Importantly, postoperative morbidity and patient-reported outcomes generally require prospective or contemporaneous standardized assessment. In our retrospective cohort, perioperative morbidity indicators and patient-reported symptoms (for example, sensory disturbance or paresthesia, pain, seroma or hematoma, and arm symptoms) were not captured in a standardized manner. Moreover, given the long enrollment period and the historical nature of many cases, these outcomes can no longer be reliably ascertained from medical records retrospectively. Therefore, a meaningful head-to-head comparison of morbidity between standard SLNB and ICBN-exposing SLNB cannot be provided, even descriptively. We believe this represents an important area for future prospective evaluation. In short, the results suggested that positive SLN location above the ICBN was a high risk factor for non-SLN metastasis. Therefore, we should not ignore ALND when positive SLNs appear in the upper group. To our knowledge, this is the first study to include the metastatic SLN location in a non-SLN metastasis prediction model.

The MSKCC nomogram is a widely recognized model for predicting non-SLN metastasis. The model includes tumor size, tumor type and grade, number of positive SLNs, number of negative SLNs, lymphovascular invasion, multifocality, and ER status. Numerous studies have validated the MSKCC model, but the AUC varies from 0.6 to 0.8 because of differences in regions and patient populations [3, 4, 17, 20, 29, 53–59]. The SCH nomogram was the first model established using a population of Chinese individuals diagnosed with breast cancer, incorporating variables such as number of positive SLNs, number of negative SLNs, lymphovascular invasion, and size of the SLN metastasis [20], with an original AUC of 0.779. Upon validation using our database, the MSKCC and SCH models demonstrated AUC values of 0.690 and 0.716, suggesting a good discriminatory ability. However, Wu et al. [29] tested the MSKCC and SCH models in 236 Chinese individuals diagnosed with breast cancer and their AUCs were only 0.677 and 0.674. The reason for this discrepancy may be the large difference in the age distribution of our patients and those of Wu et al. In our study, 56.3% (271/481) of the patients were older than 50 years, compared to 37.7% (89/236) in theirs. These findings demonstrate that the efficacy of the models is contingent upon factors such as sample size, parameter definition, parameter inclusion, and ethnic disparities, thereby limiting their applicability to breast cancer in the Chinese population.

Our model was developed in 481 SLN-positive patients and showed good discrimination (AUC 0.830), with acceptable performance in the temporal validation cohort (AUC 0.785). Compared with the MSKCC and SCH nomograms, our model demonstrated higher discrimination in our single-center Chinese cohort, possibly because it incorporates additional variables and thus better captures the local case mix. Using prespecified probability thresholds, the model identified 0.9%, 23.0%, and 27.4% of patients as candidates for omitting completion ALND under guaranteed false-negative rate targets of 0%, below 5%, and below 10%, respectively. Because AUC alone does not fully reflect clinical usefulness, we further assessed clinical utility using decision curve analysis, and our model provided greater net benefit than the MSKCC and SCH models across clinically relevant threshold probabilities. Clinically, the predicted non-SLN risk may complement Z0011 or AMAROS-based decision-making, supporting omission of completion ALND in eligible patients with low predicted risk, and informing consideration of axillary regional nodal irradiation as an alternative to ALND in patients with intermediate to high predicted risk, particularly when Z0011 or AMAROS eligibility is uncertain [10, 11].

The IBCSG23-01, AATRM, and Z0011 studies revealed no statistically significant disparities in DFS and OS between SLNB-alone and ALND groups [10, 43, 44]. Moreover, the SLNB-alone group exhibited a decrease in the incidence of postoperative upper extremity lymphedema and sensory and motor nerve injury. The AMAROS and OTOASOR studies similarly found no significant difference in prognosis between axillary radiotherapy and ALND groups, and the axillary recurrence rate was extremely low in both groups [11, 12]. Together, these findings imply that the option of disregarding ALND or selecting axillary RNI may be viable in cases of early breast cancer with a limited or moderate tumor burden in the SLNs. Current clinical trials have mostly focused on patients with 1 to 2 positive SLNs. Previous studies and our results also confirm that patients with 3 or more positive SLNs do not necessarily have non-SLN metastases, which would perhaps exempt them from ALND as well [1, 36]. Our new model predicted this subgroup and found an AUC of 0.843, which demonstrates the ability of the new model to accurately evaluate the non-SLN status of individuals diagnosed with breast cancer with 3 or more positive SLNs. Therefore, our model has the ability to precisely forecast the non-SLN status and, consequently, assess the tumor load of ALNs in Chinese women diagnosed with early breast cancer, which can provide valuable information for the subsequent decision-making regarding ALND or additional axillary RNI.

However, our study has several limitations. First, it was conducted at a single institution in China, which may limit generalizability. Although the enrollment period was long (2009–2022) and could introduce temporal heterogeneity, an era-based sensitivity analysis (2009–2014 vs. 2015–2019) demonstrated comparable discrimination with acceptable calibration (Fig. 5), thereby mitigating concerns regarding temporal performance drift. Second, metastatic SLN location is a newly proposed anatomical predictor and requires prospective multicenter validation with standardized procedures and documentation. In addition, our non–radioisotope dual-tracer approach differs from the radioisotope-based SLNB workflows commonly used in many centers, and the electronic medical record did not document the ICBN-relative location separately for ICG versus blue-stained SLNs; therefore, the transportability of this predictor to radioisotope-based settings and the concordance across tracers remain uncertain. Third, ICBN-based partitioning is operator-dependent. Although operative notes typically contained an “upper” or “lower” ICBN statement, we were unable to perform a formal inter-observer reproducibility assessment because intraoperative images and sufficiently detailed documentation for independent re-review were not routinely available. Moreover, ICBN-relative SLN location was missing in approximately 5% of otherwise eligible cases for unclear reasons, which may further limit external reproducibility and implementation. Finally, perioperative morbidity indicators and patient-reported outcomes were not collected in a standardized manner, and given the historical nature of many cases, these data could not be reliably ascertained retrospectively; thus, morbidity profiles between standard SLNB and ICBN-exposing SLNB could not be compared.

Conclusion

In this single-center retrospective cohort with temporal validation, we developed a nomogram incorporating metastatic SLN location relative to the second ICBN and standard clinicopathologic predictors to estimate the probability of non-SLN metastasis. The model showed good discrimination and calibration and provided greater net benefit than the MSKCC and SCH nomograms in our dataset. Given the operator-dependent nature of ICBN identification, differences in SLNB protocols across centers, and the lack of standardized morbidity assessment in this retrospective series, prospective multicenter validation with standardized recording is required before broader clinical adoption.

Abbreviations

SLNB

Axillary sentinel lymph node biopsy

ALN

Axillary lymph node

ALND

Axillary lymph node dissection

SLN

Sentinel lymph node

WBI

Whole-breast irradiation

RNI

Regional nodal irradiation

MSKCC

Memorial Sloan Kettering Cancer Center

SCH

Shanghai Cancer Hospital

AJCC

American Joint Committee on Cancer

ITC

Isolated tumor cells

MB

Methylene blue

ICG

Indocyanine green

ICBN

Intercostobrachial nerve

ECE

Extracapsular extension

ER

Estrogen receptor

PR

Progesterone receptor

HER-2

Human epidermal growth factor receptor 2

FISH

Fluorescent in situ hybridization

ROC

Receiver operating characteristic

DCA

Decision curve analysis

CP

cut-off point

FNR

False negative rate

DFS

Disease-free survival

OS

Overall survival

Authors’ contributions

Qunchen Zhang and Yutong Fang participated in the data analysis, organized the article writing, and critically modified the manuscript. Bingfeng Chen1, Lifang He, Chunfa Chen Guangsheng Huang modified the manuscript, drafted the manuscript and were responsible for the acquisition of data; Cuiping Guo, Zexiao Chen, Huancheng Zeng, Yongqu Zhang, and Jundong Wu contributed to the literature search, and correct language expression. All authors read and approved the manuscript and agree to be accountable for all aspects of the research in ensuring that the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Funding

This work was supported by funds from the 2023 Science and Technology Innovation Strategy Project of Guangdong Province (Big Project + Task List), China (No. STKJ2023009,20230403); funds from the Bethune“Wings for Research”-Supporting Research Projects in the Field of Therapeutics, China (No. 2024-YJ-154-S-016); the Young Research Fund Project of the Cancer Hospital of Shantou University Medical College (No. 2023A002); Jiangmen Central Hospital Foundation for Distinguished Young Scholars (No. J202404); and the Science and Technology Project of Shantou, China (No. 200601155261085).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of the Cancer Hospital of Shantou University Medical College (NO. 2023086), and informed consents were obtained from all patients.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.